The Future of AI in Games: Marek Rosa at Game Developer Session 2022

How to leverage the power of large language models (LLMs) for game development and games. Taking examples from GoodAI's in-the-works AI Game, Marek Rosa demonstrates the benefits and potential of utilizing an AI-driven approach to NPCs for richer game personalities, complex interactions, and more emergent gameplay.
Presented at the Game Developers Session Conference
Date: September 24, 2022
Timestamps:
2:25 Large language Models
4:44 Limitations of LLMs
5:35 Cost of Inference
5:55 Hallucinations
6:50 Statelessness of LLMS - memory loss
7:38 Observation space
8:19 When things don't work
9:18 Why we want evil AI
10:55 Evil AI examples
11:38 Grounding LLM in the game world
12:10 Optimizing LLM for more
13:40 Emerging complexity
14:14 Self-aware NPCs
16:26 Last 10 years in AI: to 2022 stable diffusion
19:10 Image generators
23:20 Predictions for the Next 5-10 years
25:10 New research on memory improvement of LLMs
30:10 What is AGI?
33:15 AI in games as an incremental path towards AGI
41:35 Q&As
***
We're hiring! We are always looking for new colleagues who seek technical and artistic excellence and who like to push their limits. Check out our links below!
Website: www.goodai.com
Career: www.goodai.com/careers/
Twitter: GoodAIdev
Facebook: GoodArtificialIntelligence
Linkedin: www.linkedin.com/company/goodai/

[@intergalacticGM]

This was somewhat interesting how machine learning is applied to games this time. I was expecting something different, like a custom built algorithm. And more related to strategy like application, like the one used to beat dota or starcraft human champions. Also I doubt anything mentioned here will be applied to space engineers soon. So I guess state machines are the thing we'll most likely get with the next update of the game.

[@odiihinia]

Use Elevenlab's voice generative AI to fully integrate voice acting into generated text.